Audio File MCP App

What the Audio File MCP App MCP server does

Audio File MCP App MCP server opens a local audio file inside an MCP host and renders an in-conversation interface for listening and visual inspection. The interface combines playback controls with a waveform, a reassigned spectrogram, file metadata, and loudness measurements. It is intended for audio review tasks where the model needs access to the state of what the user is hearing or examining.

The waveform uses spectral-centroid-based colouring to give a visual indication of tonal balance across the file. The spectrogram uses logarithmic frequency bins starting at 20 Hz and a reassigned time-frequency display, which helps show transients and tonal components with more detail than a basic STFT view.

How it works

The server runs locally over stdio. A host calls the display_audio_file tool with an absolute local path, such as /Users/me/Music/track.wav. The tool then renders the audio UI in the host. No remote audio service or external API is described in the provided setup.

During playback or cursor movement, the interface exposes state to the model. This includes metadata, global loudness statistics, the current playhead position, and any selected region. Loudness calculations include integrated loudness, loudness range, true peak, sample peak, and RMS. Momentary and short-term LUFS, along with cursor-position sample peak and RMS, are also available while playing or hovering.

Dragging across the timeline creates a selected region that playback loops. Its start and end positions are passed to the model with the rest of the audio state. This makes it possible to refer to a particular section during a follow-up request.

Tools and capabilities

The Audio File MCP App MCP server exposes the display_audio_file tool. Its supported inputs and views include:

• A local audio file path.
• Optional timeline annotations supplied directly as an object.
• Optional annotations loaded from a JSON file through annotationsPath.
• Waveform, playback, metadata, loudness, and spectrogram views.
• Looping selection of a time range.
• Annotation lanes with labels, colours, time spans, and optional opacity envelopes.

Annotation times use seconds on the audio file’s own timeline. Lanes can mark sections, events, or results from another analysis. The model receives lanes active at the playhead, as well as lanes starting, ending, or active within a selected region.

Setup and configuration

Install the server with the command npx -y @counterpoint-studio/audio-file-mcp-app. It can be added to Claude Desktop, Codex Desktop, Visual Studio Code, or Goose using each host’s MCP configuration flow. Claude Desktop also supports installing the latest .mcpb bundle from the project’s releases. MCP Inspector can launch the same stdio command for testing.

The project’s development instructions use pnpm. They include commands for installing dependencies, building DSP components, running the server with tsx, and running tests. The DSP build requires emsdk. No environment variables are specified for normal operation.

Limitations and notes

The server is designed for local files and uses a stdio transport. The provided material does not describe remote file access, editing or exporting audio, recording, or audio transformation tools. The README identifies compatibility with Claude Desktop, Codex Desktop, Visual Studio Code, Goose, and MCP Inspector; support for other hosts is not established here.

Annotation payloads can become large, so annotationsPath is available for a JSON file with the same lanes structure. Within one lane, overlapping spans are truncated at the next span’s start so the rendered rows do not overlap.